◈AI Codex
Foundation Models & LLMsRole-Specific

How to think about Claude's context window (and when it actually matters)

In brief

A million tokens sounds like more room than anyone could use. In practice, how you use that space — not whether you run out of it — is what changes the quality of your outputs.

4 min read·Context Window

Contents

♡Sign in to save

Claude's context window — 1,000,000 tokens on Claude Fable 5, Claude Opus 5 and Claude Sonnet 5, roughly 555,000 words — is large enough that most people never come near the limit. But "won't run out of space" and "using the space well" are different things, and the gap between them got wider as the window got bigger.

Here's how to actually think about it.

The whiteboard metaphor

The context window is like a whiteboard Claude can see during a conversation. Everything on the whiteboard — your instructions, the conversation so far, documents you've pasted in, Claude's previous responses — is available to Claude while it's answering.

When the whiteboard fills up, old content gets erased from the top. Claude does have memory that carries between conversations, but that is a separate mechanism — retained entries that get loaded onto the whiteboard, not a bigger whiteboard. Within a conversation, what is on the board is what Claude is working from.

This has two practical implications.

Long conversations drift

In a long back-and-forth, the instructions you gave at the start start to feel further away. Not because Claude "forgets" them exactly, but because there's now a lot of other content competing for attention in the same space.

If you notice that Claude's behaviour starts to drift across a long conversation — becoming less precise, reverting to generic patterns — this is often why. Solutions: start a fresh conversation with your key instructions repeated, or use Projects to keep instructions persistent and separate from the conversation itself.

Large documents need care

If you paste a 50-page document and then ask Claude questions about it, Claude has the whole document. But the quality of answers depends on how clearly the relevant section stands out from the noise.

For documents with lots of sections, consider telling Claude where to look: "Focus on the section titled 'Pricing' when answering this question." This isn't a workaround — it's good practice. You wouldn't give a colleague a 50-page document and ask them a question without pointing them to the right page.

When the context window actually matters for operators

Two scenarios where the context window becomes a real consideration:

Building with the API. If you're running Claude in a product where context accumulates across many turns, you need to think about context management — when to summarise, when to reset, what to keep. This is a technical design decision.

Very long analysis tasks. Asking Claude to read and synthesise a whole book, a full year of customer feedback, or a large codebase — these tasks benefit from thinking about how you structure the input, not just dumping everything in at once.

For most conversational use, the context window is something you can ignore. When it becomes relevant, these are the patterns to know.

Further reading

Official training on this: AI Capabilities and Limitations (13 lessons · 3.5 hr) on Claude Academy, free.

Weekly brief

For people actually using Claude at work.

Each week: one thing Claude can do in your work that most people haven't figured out yet — plus the failure modes to avoid. No tutorials. No hype.

No spam. Unsubscribe anytime.

What to read next

All articles →
Context Window·Core Definition·5 min

The whiteboard every AI conversation shares

Context window is the single number that shapes everything about how Claude thinks with you — and most people are using only a fraction of it.

Context Window·How It Works·5 min

The context window in practice: what it means for how you work

The context window shapes what Claude can and can't do in any given conversation. Here is how to work with it.

How It Works·14 min

Claude Fable 5.1: cache reads at a quarter of the price, and three changes that break existing code

On September 1, 2026 Anthropic shipped Claude Fable 5.1 at the same $10/$50 per million tokens as Fable 5, with cache reads cut from $1.00 to $0.25 per million. That single line makes long agent sessions roughly 25% cheaper, and up to 45% cheaper for heavily agentic work. Three things break on migration: forced tool use, cross-model thinking blocks, and any code that edits earlier turns. Here is what changed and what to check before you switch the model ID.